基于图像序列的多乘客目标识别和计数的研究
|
摘要
自动乘客计数(Automatic Passenger Counting,APC)技术是不需要人工干预,通过一定的检测方法,对一定场所内人流大小的统计。自动乘客计数尤其是现代化智能公交系统的一个重要功能模块,为公交实时调度、优化和预测提供决策依据。优秀的客流计数系统能够精确记录乘客流量,节约商业运作成本,便于管理者实时了解运营情况,实现更高效的公交系统的运营管理。
相比于其它自动乘客计数系统,基于数字图像处理技术的自动乘客计数方法日趋成为现代化智能交通系统领域的研究热点问题。其中,多运动目标检测与计数的功能设计是该研究的核心。多运动目标检测与计数的结果直接影响整个系统精度。
本文采用数字图像处理平台,提出了一种新的基于图像序列的多乘客目标识别算法,构建以摄像头采集的实时图像为输入的自动乘客计数系统。通过自动图像识别手段,系统自动输出上下车乘客人数。多乘客目标识别需要同时实现多乘客目标检测、多乘客目标跟踪、多乘客目标计数这三个部分的功能。光照、人头遮挡和天气变化等因素会影响系统功能的实现。为提高图像检测的鲁棒性,本文以自适应的模型方法为背景图像进行建模。在乘客目标识别方面,利用人头与圆形之间的相似度构造评价模型,减小系统计算量,提高效率。在目标跟踪方法的选择上,利用包含颜色信息和概率特征的Mean Shift算法设计仿射变换,提高系统的鲁棒性。计数方法的研究中,提出目标序列概念,实现对公交车运动目标的跟踪和计数。
本文对基于图像序列的多乘客目标识别和计数中的新技术、新方法进行实验仿真。通过实验表明,系统能够有效掌控不同时间不同区域的客流信息,实现对上下车客流量的准确统计。从而,本研究为智能交通系统提供科学依据。
Automatic Passenger counting (APC) technology can automatically count the number of the passengers at some districts through using some testing method. Automatic passenger counting is especially one of important function module of modern intelligent bus system, it can provide insurance for decision-making of real-time scheduling, optimization and forecasting. Excellent passengers counting system can precisely keep track of passengers’number, save business operation cost, assist managers to understand the operation situation in real time, achieving a more efficient public transport system operational management.
In comparison with other passengers’automatic counting system, an automatic passenger count method based on digital image processing technology is increasingly to be a focus in the field of modern intelligent transportation system research. Among lots of methods, the design of multi-moving targets detection and counting functions is the very core content. The effect of multi-moving targets detection and counting directly affect the entire system’s precision.
This paper uses the digital image processing platform, puts forward a new algorithm about multi-passengers target recognition based on image sequence, and constructs an automatic passengers counting system whose input is the real-time images collected by cameras. Through the automatic image recognition method, the system can automatically output the passengers’number. Multi-passengers target recognition needs to realize these three functions: multi- passengers target detection, multi- passengers target tracking, multi- passengers target counting. The existence of light, head sheltering, changes of the weather changes can all influence the realization of the function of the system. In order to improve the robustness of the image detection, background image modeling is based on adaptive model methods. About the passenger target recognition, it sets up an evaluation model using the similarity between shape features of the passengers’heads and the rounds, which reduces the system calculation and improves the efficiency; About targets tracking algorithm, Mean Shift algorithm including color and probability information is used to design affine transformation, improving the system‘s robustness. About counting methods, target sequence concept is put forward to assist realizing the tracking and counting of the moving targets on bus.
This paper includes some simulation experiences about the new technology, new method in multi-passengers target recognition and counting field based on image sequence. The experiments show that the system can effectively control information of different times and different areas about passengers flow, realizes its accurate traffic statistics. Thus, researches in this paper provide intelligent transportation system.
相比于其它自动乘客计数系统,基于数字图像处理技术的自动乘客计数方法日趋成为现代化智能交通系统领域的研究热点问题。其中,多运动目标检测与计数的功能设计是该研究的核心。多运动目标检测与计数的结果直接影响整个系统精度。
本文采用数字图像处理平台,提出了一种新的基于图像序列的多乘客目标识别算法,构建以摄像头采集的实时图像为输入的自动乘客计数系统。通过自动图像识别手段,系统自动输出上下车乘客人数。多乘客目标识别需要同时实现多乘客目标检测、多乘客目标跟踪、多乘客目标计数这三个部分的功能。光照、人头遮挡和天气变化等因素会影响系统功能的实现。为提高图像检测的鲁棒性,本文以自适应的模型方法为背景图像进行建模。在乘客目标识别方面,利用人头与圆形之间的相似度构造评价模型,减小系统计算量,提高效率。在目标跟踪方法的选择上,利用包含颜色信息和概率特征的Mean Shift算法设计仿射变换,提高系统的鲁棒性。计数方法的研究中,提出目标序列概念,实现对公交车运动目标的跟踪和计数。
本文对基于图像序列的多乘客目标识别和计数中的新技术、新方法进行实验仿真。通过实验表明,系统能够有效掌控不同时间不同区域的客流信息,实现对上下车客流量的准确统计。从而,本研究为智能交通系统提供科学依据。
Automatic Passenger counting (APC) technology can automatically count the number of the passengers at some districts through using some testing method. Automatic passenger counting is especially one of important function module of modern intelligent bus system, it can provide insurance for decision-making of real-time scheduling, optimization and forecasting. Excellent passengers counting system can precisely keep track of passengers’number, save business operation cost, assist managers to understand the operation situation in real time, achieving a more efficient public transport system operational management.
In comparison with other passengers’automatic counting system, an automatic passenger count method based on digital image processing technology is increasingly to be a focus in the field of modern intelligent transportation system research. Among lots of methods, the design of multi-moving targets detection and counting functions is the very core content. The effect of multi-moving targets detection and counting directly affect the entire system’s precision.
This paper uses the digital image processing platform, puts forward a new algorithm about multi-passengers target recognition based on image sequence, and constructs an automatic passengers counting system whose input is the real-time images collected by cameras. Through the automatic image recognition method, the system can automatically output the passengers’number. Multi-passengers target recognition needs to realize these three functions: multi- passengers target detection, multi- passengers target tracking, multi- passengers target counting. The existence of light, head sheltering, changes of the weather changes can all influence the realization of the function of the system. In order to improve the robustness of the image detection, background image modeling is based on adaptive model methods. About the passenger target recognition, it sets up an evaluation model using the similarity between shape features of the passengers’heads and the rounds, which reduces the system calculation and improves the efficiency; About targets tracking algorithm, Mean Shift algorithm including color and probability information is used to design affine transformation, improving the system‘s robustness. About counting methods, target sequence concept is put forward to assist realizing the tracking and counting of the moving targets on bus.
This paper includes some simulation experiences about the new technology, new method in multi-passengers target recognition and counting field based on image sequence. The experiments show that the system can effectively control information of different times and different areas about passengers flow, realizes its accurate traffic statistics. Thus, researches in this paper provide intelligent transportation system.
引文
[1]李苗,刘卫宁,孙棣华,一种适于公交乘客计数的自适应背景更新算法[ J],计算机工程与应用,2006(36):216~218
[2]赵敏,基于运动目标识别的自动乘客计数技术研究[ D],重庆:重庆大学,2006
[3]钟伟,余松煜,丙雨,图像序列处理中的形态分割方法,上海:上海交通大学学报,2001,9 P,13~14
[4]陈邦忠.红外图像序列的运动目标检测与运动分析[ D],北京:清华大学,2002
[5]王建明,多目标模糊识别优化决策理论与应用研究,[大连理工大学博士学位论文],2004
[6]孙兆林,MATLAB6X图像处理,北京:清华大学出版社,2002
[7]沈庭芝,方子文,数字图像处理及模式识别[M],北京:北京理工大学出版社,2000
[8]杨国栋,江海宗,ITS——智能交通[J],智能建筑与城市信息,2008,142(9):9~12
[9] Dan Kong, Doug Gray, Hai Tao. A viewpoint invariant approach for crowd counting [A].In: Proceedings of the 18th International Conference on Pattern Recognition[C]. Hong Kong, China,2006,1187~1190
[10] Marana A N. Real-time crowd density estimation using images [J].Advances in visual computing, 2005, 3804.355~362
[11]于海滨,付伟,刘济林,视觉客流检测中的动态轮廓匹配方法[J],浙江大学学报(工学版),2008,42(3):412~417
[12]王成儒,顾广华,一种采用背景统计技术的视频对象分割算法[J],光电工程,2004,31(8):57~60
[13]于海滨,刘济林,基于区域视差提取的视觉客流统计方法[J],传感技术学报,2007,20(7):1546~1550
[14]Rabaud V, Belongie S. Counting crowded moving objects [A]. In: Proceedings of the IEEE Computer Society Conference on Computer Vision and patlern Recognition[C].NewYork, USA,2006,705~711
[15]张会军,基于图像的动目标检测技术[J],微计算机信息,2007,8(1):299~300
[16]朱晓宏,公交客流信息采集的方法与技术[ J],城市公共交通, 2007(7):20~21
[17]Bastian Leibe, Edgar Seemann,Bernt Schiele. Pedestrian detection in crowded scenes[A].In: Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition[c],San Diego, CA, USA, 2005,878~885
[18]于成忠,朱骏,袁晓辉,基于背景差法的运动目标检测[J],东南大学学报,2005,35(11):159~161
[19]汪亚明,基于动态图像序列的自动扶梯客流量的测量[J],华北工学院测试技术学报,2002(1):14~16
[20]赵敏,基于运动目标识别的自动乘客计数技术研究[D],重庆大学学报,2006
[21]叶青,周文远,宋宇等,一种实时视频采集处理系统的设计与实现[J],光电子激光,2006,1(17):42~46
[22] Hartley R,Zissennan A.Multiple view geometry in computer vision[M].Cambridge:Cambridge University Press,2000
[23] Castleman K R.Digital image processing[M].Beijing:Tsinghua University Press,1998,492~495
[24] Varanasi M K,Aazhang B.Near-Optimum Detection in Synchronous Code Division Multiple Acess System[J].IEEE Trans.on Communications.2001,39:725~736
[25] J.H.Friedman,J.L.Bentley,R.A.Finkel.An algorithm for finding best matches in logarithmic expected time.ACM Transactions on Mathematical Software.1977,3(3):209~226
[26] Jeffrey S.Bei s,D.G.Lowe.Shape indexing using approximate nearestneighbour search in high dimensional spaces[C].In Conference on Computer Vision and Pattern Recognition,Puerto Rico,1997,1000~1006
[27] Tanimizu K Meguro A,Ishii A.High speed defectdetection method for color printedmatter[C]//Pro-ceedings of Industrial Electronics Conference.Los An-geles,USA:IEEE.1990,653~658
[28] David A.Forsyth,Jean Ponce.A Modern Approach.Computer Vi sion,2003
[2]赵敏,基于运动目标识别的自动乘客计数技术研究[ D],重庆:重庆大学,2006
[3]钟伟,余松煜,丙雨,图像序列处理中的形态分割方法,上海:上海交通大学学报,2001,9 P,13~14
[4]陈邦忠.红外图像序列的运动目标检测与运动分析[ D],北京:清华大学,2002
[5]王建明,多目标模糊识别优化决策理论与应用研究,[大连理工大学博士学位论文],2004
[6]孙兆林,MATLAB6X图像处理,北京:清华大学出版社,2002
[7]沈庭芝,方子文,数字图像处理及模式识别[M],北京:北京理工大学出版社,2000
[8]杨国栋,江海宗,ITS——智能交通[J],智能建筑与城市信息,2008,142(9):9~12
[9] Dan Kong, Doug Gray, Hai Tao. A viewpoint invariant approach for crowd counting [A].In: Proceedings of the 18th International Conference on Pattern Recognition[C]. Hong Kong, China,2006,1187~1190
[10] Marana A N. Real-time crowd density estimation using images [J].Advances in visual computing, 2005, 3804.355~362
[11]于海滨,付伟,刘济林,视觉客流检测中的动态轮廓匹配方法[J],浙江大学学报(工学版),2008,42(3):412~417
[12]王成儒,顾广华,一种采用背景统计技术的视频对象分割算法[J],光电工程,2004,31(8):57~60
[13]于海滨,刘济林,基于区域视差提取的视觉客流统计方法[J],传感技术学报,2007,20(7):1546~1550
[14]Rabaud V, Belongie S. Counting crowded moving objects [A]. In: Proceedings of the IEEE Computer Society Conference on Computer Vision and patlern Recognition[C].NewYork, USA,2006,705~711
[15]张会军,基于图像的动目标检测技术[J],微计算机信息,2007,8(1):299~300
[16]朱晓宏,公交客流信息采集的方法与技术[ J],城市公共交通, 2007(7):20~21
[17]Bastian Leibe, Edgar Seemann,Bernt Schiele. Pedestrian detection in crowded scenes[A].In: Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition[c],San Diego, CA, USA, 2005,878~885
[18]于成忠,朱骏,袁晓辉,基于背景差法的运动目标检测[J],东南大学学报,2005,35(11):159~161
[19]汪亚明,基于动态图像序列的自动扶梯客流量的测量[J],华北工学院测试技术学报,2002(1):14~16
[20]赵敏,基于运动目标识别的自动乘客计数技术研究[D],重庆大学学报,2006
[21]叶青,周文远,宋宇等,一种实时视频采集处理系统的设计与实现[J],光电子激光,2006,1(17):42~46
[22] Hartley R,Zissennan A.Multiple view geometry in computer vision[M].Cambridge:Cambridge University Press,2000
[23] Castleman K R.Digital image processing[M].Beijing:Tsinghua University Press,1998,492~495
[24] Varanasi M K,Aazhang B.Near-Optimum Detection in Synchronous Code Division Multiple Acess System[J].IEEE Trans.on Communications.2001,39:725~736
[25] J.H.Friedman,J.L.Bentley,R.A.Finkel.An algorithm for finding best matches in logarithmic expected time.ACM Transactions on Mathematical Software.1977,3(3):209~226
[26] Jeffrey S.Bei s,D.G.Lowe.Shape indexing using approximate nearestneighbour search in high dimensional spaces[C].In Conference on Computer Vision and Pattern Recognition,Puerto Rico,1997,1000~1006
[27] Tanimizu K Meguro A,Ishii A.High speed defectdetection method for color printedmatter[C]//Pro-ceedings of Industrial Electronics Conference.Los An-geles,USA:IEEE.1990,653~658
[28] David A.Forsyth,Jean Ponce.A Modern Approach.Computer Vi sion,2003